Seasonal variation in incidence of acute myocardial infarction and cardiovascular disease risk factors in a subarctic population : the Tromsø Study

Seasonal variation in

acute myocardial infarction and cardiovascular disease risk factors

in a subarctic population

The Tromsø Study

Laila Arnesdatter Hopstock

Department of Community Medicine University of Tromsø

Tromsø, Norway

2012

“Whoever wants to pursue properly in the science of medicine must proceed thus. First he ought to consider what effects each season of the year can produce; for the seasons are not all alike, but differ widely both in themselves and at their changes.”

Hippocrates

ACKNOWLEDGEMENTS

This research was performed at the Department of Community Medicine (ISM) at the University of Tromsø, Norway and at the Institute of Health and Biomedical Innovation (IHBI) at the Queensland University of Technology, Australia, from October 2007 to June 2012. During this time, many great people have followed me on this path to a PhD. Thanks to all of you! In particular;

Thanks to my co-supervisor Kaare H. Bønaa for giving me the chance to work with this interesting and challenging research idea, for helpful guidance and advice. Thanks also to my co-supervisor Inger Njølstad for your support and inspiration for me as a young researcher walking my first steps into the world of epidemiological research. I am deeply grateful to my main supervisor Tom Wilsgaard, for following me throughout my time as a PhD student. You have been extremely patient in spite of my struggle with not just advanced but also basic statistical challenges – and even simple mathematical truths. Thanks for always being there, sharing your vast knowledge and being enthusiastic about new ideas.

Thanks must also go to my co-authors for your in-depth knowledge in various parts of science. To Ane Schwencke Fors for introducing me to the mysteries of meteorology, to Ellisiv B. Mathiesen and Maja-Lisa Løchen for your valuable contributions as experts in cardiovascular disease epidemiology, to Adrian G.

Barnett for welcoming me to the world of advanced methods for analysing seasonality, and to Jan Mannsverk for your updated clinical expertise in the field of cardiovascular disease. Thanks to all of you for many important discussions of study limitations and impacts of results.

I also want to thank all my dear colleagues at ISM and the members of the Health Research Group at IHBI for being such a creative and helpful learning and working environment. I send special thanks to my friends and colleagues from the EPINOR PhD School. How had life as a PhD student been without you?

We’ve shared ups and downs, discussed all kinds of epidemiological topics and had fabulous parties in and outside the office.

The warmest thanks to my close friends and family! Thanks to my best friend and nurse colleague Kirsten for never letting me forget about the clinical world – the real world – where we meet the patient with the heart disease. Thanks to my helpful and always positive mother who took position as babysitter so that I could finish the last part of the thesis in due time. And to my dear husband Fredrik and my baby daughter Liv Elvina; Mina älsklingar! Thanks for all your patience and support.

Finally, thanks to all the participants in the Tromsø Study for sharing your blood, valuable time and bodily secrets, contributing to what is a huge collection of valuable data material making this research possible.

Laila Tromsø, June 2012

1 CONTENTS

ACKNOWLEDGEMENTS

SUMMARY ... 3

SAMMENDRAG ... 4

LIST OF PAPERS ... 5

ABBREVIATIONS ... 6

LIST OF TABLES AND FIGURES ... 7

1. INTRODUCTION – A REVIEW OF THE LITERATURE ... 8

1.1A HISTORICAL PERSPECTIVE ... 8

1.2SEASONAL VARIATION – THE BURDEN OF DISEASE ... 8

1.3A SYSTEMATIC REVIEW ... 9

1.3.1 Objectives of the review ... 9

1.3.2 The search strategy ... 9

1.3.3 Inclusion and exclusion criteria ... 10

1.3.4 Procedure ... 10

1.3.5 Results ... 11

1.3.5.1 Seasonal variation in myocardial infarction morbidity and mortality ... 11

1.3.5.2 Effect of weather on myocardial infarction morbidity and mortality ... 11

1.3.5.3 Seasonal variation in cardiovascular disease risk factors... 12

1.3.5.4 Methodological considerations – the review ... 12

1.3.5.5 Geographical differences – the review ... 13

1.3.5.6 Limitations of the review... 14

1.3.6 Discussion with rationale for the thesis ... 15

2. AIMS OF THE THESIS ... 16

3. METHODS ... 17

3.1TROMSØ – LOCATION, CLIMATE AND SEASONS ... 17

3.1.1 Meteorological data and definitions of seasons used in paper 1-3 ... 17

3.2STUDY POPULATION –THE TROMSØ STUDY ... 18

3.2.1 Study population used in paper 1–3 ... 19

3.3MYOCARDIAL INFARCTION CASE IDENTIFICATION AND DEFINITION USED IN PAPER 1 AND 2 ... 20

3.4CARDIOVASCULAR DISEASE RISK FACTORS MEASUREMENTS USED IN PAPER 3 ... 20

3.5DATA FROM QUESTIONNAIRES USED IN PAPER 1 ... 21

3.6STATISTICAL ANALYSES ... 21

3.6.1 Cosinor ... 21

3.6.2 Statistical models used in paper 1-3 ... 22

4. RESULTS – SUMMARY OF PAPERS ... 25

4.1PAPER 1 ... 25

4.2PAPER 2 ... 25

4.3PAPER 3 ... 26

2

5. DISCUSSION – METHODOLOGY ... 27

5.1OVERALL CONSIDERATIONS OF INTERNAL VALIDITY ... 27

5.1.1 Study design ... 27

5.1.2 Study population – selection bias, non-response bias and loss to follow-up ... 27

5.1.3 Measurement of outcome – misclassification bias ... 27

5.1.4 Measurement of exposure – construct validity ... 28

5.1.5 Measurement of exposure – misclassification bias ... 28

5.1.6 Statistical methods – validity ... 28

5.1.7 The preconception of the researcher – bias in analysis, interpretation and presentation .. 29

5.2PAPER 1 ... 29

5.2.1 Bias in analysis, interpretation and presentation ... 29

5.3PAPER 2 ... 29

5.3.1 Measurement of exposure – misclassification bias ... 29

5.3.2 Confounding ... 30

5.4PAPER 3 ... 30

5.4.1 Study design ... 30

5.4.2 Measurement of outcome – measurement error and biological variation ... 30

5.5EXTERNAL VALIDITY ... 31

6. DISCUSSION - RESULTS ... 32

6.1SEASONAL VARIATION – GEOGRAPHICAL DIFFERENCES ... 32

6.2THE BIOLOGY... 33

6.2.1 Physiological and pathophysiological responses to season and weather ... 33

6.2.1.1 Association and causality ... 34

6.2.2 Adaptation and acclimatization ... 34

6.3THE CULTURE ... 35

6.3.1 Housing ... 35

6.3.2 Clothing ... 36

6.3.3 Outdoor exposure ... 37

6.3.4 Living standard ... 38

6.4SEASONAL VARIATION – SEX AND AGE DIFFERENCES ... 38

7. CONCLUSIONS AND FUTURE PERSPECTIVES ... 39

7.1TODAY ... 39

7.1.1 Living in the High North ... 39

7.2TOMORROW ... 39

8. TABLES AND FIGURES... 40

9. REFERENCES ... 77 EPILOGUE

APPENDIX PAPER 1-3

3 SUMMARY

A seasonal pattern with winter peak in acute myocardial infarction incidence and cardiovascular disease risk factors is observed in studies worldwide. However, several previous studies have

methodical limitations and few are performed in cold climate areas. The aim of this thesis is to assess the effect of season and meteorological factors on first-ever myocardial infarction and the seasonal variation in cardiovascular disease risk factors in a subarctic adult population with long-term follow- up, using appropriate methods with adjudicated outcomes and well-defined exposures.

The population-based Tromsø Study consists of more than 40,000 individuals living in a subarctic climate in Northern Norway. The cohort members have been examined up to nine times in six repeated health surveys in the years between 1974 and 2008. Data on myocardial infarction and risk factors have been collected throughout follow-up. The thesis consists of three studies. The first study is an analysis of the seasonal variation in fatal and non-fatal incident myocardial infarction. The second study is an analysis of the effect of temperature, wind, atmospheric pressure, humidity and snowfall on incident myocardial infarction. The third study is an analysis of the seasonal variation in systolic and diastolic blood pressure, heart rate, body weight, total cholesterol, triglycerides, high density lipoprotein cholesterol, C-reactive protein and fibrinogen.

The results from the present studies show that there is a seasonal variation with increased risk of myocardial infarction in the darkest winter months, however small compared to what has been observed in other populations, especially those in warmer climates. There is little effect of weather variables, although cold temperatures and heavy snowfall may increase the risk of myocardial infarction among older individuals. Cardiovascular disease risk factors have a seasonal pattern, although the sizes of the seasonal changes are likely too small to contribute to acute cardiovascular disease events.

The findings implies that, compared to populations in warmer climates, this subarctic populations is little effected by season and weather, probably due to long-term adaption to a harsh climate, mainly through behavioural protection.

4 SAMMENDRAG

Årstidsvariasjon med topp om vinteren i akutt hjerteinfarkt insidens og risikofaktorer for hjertekarsykdom er observert i studier verden over. Mange tidligere studier har metodiske

begrensninger og få er utført i områder med kaldere klima. Hensikten med denne avhandlingen er å undersøke effekt av årstid og meteorologiske faktorer på insidens av førstegangs hjerteinfarkt samt årstidsvariasjon i risikofaktorer for hjertekarsykdom i en subarktisk befolkning med lang

oppfølgingstid, ved bruk av adekvate metoder med validerte endepunkt og veldefinerte eksponeringsvariabler.

Den befolkningsbaserte Tromsøundersøkelsen består av over 40,000 mennesker bosatt i et

subarktisk klima i Nord-Norge. Studiedeltagerne har blitt undersøkt opp til ni ganger i seks repeterte helseundersøkelser i årene mellom 1974 og 2008, der det har blitt samlet inn informasjon om hjerteinfarkt og risikofaktorer. Avhandlingen består av tre studier. Den første studien er en analyse av årstidsvariasjon i insidens av fatale og ikke-fatale hjerteinfarkt. Den andre studien er en analyse av effekt av temperatur, vind, atmosfærisk trykk, fuktighet og snøfall på hjerteinfarktinsidens. Den tredje studien er en analyse av årstidsvariasjon i systolisk og diastolisk blodtrykk, hjertefrekvens, kroppsvekt, totalkolesterol, triglyserider, high density lipoprotein kolesterol, C-reaktivt protein og fibrinogen.

Resultatene fra disse studiene viser at det er en årstidsvariasjon med økt risiko for førstegangs hjerteinfarkt i de mørkeste vintermånedene, men at denne er liten sammenlignet med økningen om vinteren sett i land med varmere klima. Effekten av værvariabler er generelt liten, men hos eldre er det økt risiko for førstegangs hjerteinfarkt ved kulde og etter store snøfall. Risikofaktorer for hjertekarsykdom har et årstidsmønster, men størrelsen på variasjonen gjennom året er liten og vil lite trolig kunne bidra til akutt hjertekarsykdom.

Funnene viser at denne subarktiske befolkningen er lite påvirket av årstid og vær sammenlignet med befolkninger i varmere områder, antagelig på grunn av langtids adapsjon til et barskt klima,

hovedsakelig via beskyttende atferd.

5 LIST OF PAPERS

This thesis is based on the following three papers, referred to in the text as paper 1, 2 and 3.

Paper 1

Hopstock LA, Wilsgaard T, Njølstad I, Mannsverk J, Mathiesen EB, Løchen ML, Bønaa KH. Seasonal variation in incidence in acute myocardial infarction in a sub-Arctic population: the Tromsø Study 1974–2004. European Journal of Cardiovascular Prevention and Rehabilitation 2011;18(2):320-325.

Paper 2

Hopstock LA, Fors AS, Bønaa KH, Mannsverk J, Njølstad I, Wilsgaard T. The effect of daily weather conditions on myocardial infarction incidence in a subarctic population: the Tromsø Study 1974–

2004. Journal of Epidemiology and Community Health 2012;66:815-820.

Paper 3

Hopstock LA, Barnett AG, Bønaa KH, Mannsverk J, Njølstad I, Wilsgaard T. Seasonal variation in cardiovascular disease risk factors in a subarctic population: the Tromsø Study 1979–2008. Journal of Epidemiology and Community Health 2012; doi 10.1136/jech-2012-201547 [Online August 2, 2012].

6 ABBREVIATIONS

A autumn

ACS acute coronary syndrome AF atrial fibrillation

AP angina pectoris

ASAT aspartate aminotransferase ASHD arteriosclerotic heart disease

°C degrees Celsius

CA cardiac arrest

CAD coronary artery disease CCU coronary care unit

CD coronary disease

CHD coronary heart disease CICU coronary intensive care unit CK/CK-MB creatine kinase

CRP C-reactive protein CVD cardiovascular disease DBP diastolic blood pressure ECCO echocardiography ECG electrocardiogram g/L gram per litre GP general practitioner

HF heart failure

ICD international classification of diseases

IHD ischemic heart disease HDL high-density lipoprotein

HR heart rate

LD lactate dehydrogenase LIA limited available information

kg kilogram

MeSH Medical Subject Headings MI myocardial infarction mmHg millimetre mercury mmol/L millimol per litre

NIA no information available PCI percutaneous coronary

intervention

ROSC return of spontaneous circulation

SBP systolic blood pressure SCD sudden cardiac death

Sp spring

Su summer

SV seasonal variation

UV ultraviolet

W winter

WHO World Health Organization

yrs years

7 LIST OF TABLES AND FIGURES

Table 1.

Seasonal variation in myocardial infarction. Description of studies 1946–2007. Page 43

Table 2.

Weather effects on myocardial infarction. Description of studies 1946–2007. Page 55

Table 3.

Seasonal variation in cardiovascular disease risk factors. Description of studies 1946–2007. Page 65

Table 4.

Survey number, year of screening, number of participants and attendance rate by gender,

and birth year. The Tromsø Study 1974–2008. Page 77

Table 5.

Participants by number of examinations attended. The Tromsø Study 1974–2008. Page 79

Figure 1.

Flowchart of study samples, papers 1–3. The Tromsø Study. Page 81

Figure 2.

The Cosinor curve with parameters. Page 83

8 1. INTRODUCTION – A REVIEW OF THE LITERATURE

This thesis includes an introduction to the theme seasonal variation and weather effects on morbidity and mortality from myocardial infarction (MI), and seasonal variation in cardiovascular disease (CVD) risk factors, based on a systematic review and a discussion of the findings pointing out the rationale for the thesis. The aims, methods and results of the thesis will then be presented, before the study limitations and study findings will be discussed, ending with future perspectives for this field of research.

1.1 A historical perspective

The knowledge of effects of season and weather on human health has probably existed in all cultures throughout history. The first known written source describing this is from Hippocrates ¹. In his work

“Airs Water Places” he stresses the importance of recognizing that geographical conditions and climate influence health. The work is partly medical and partly ethnographical ¹, describing climate- and season-related diseases among populations living in different geographical areas. In the Nordic countries, the Swedish astronomer Pehr Wargentin was the first to describe the seasonal pattern in mortality with his report from 1767 ². Increased winter mortality from “acute coronary occlusion”

was first described in 1926 ³ whereas the relationship between “coronary artery thrombosis” and meteorological variables was first described in 1938 ⁴. Seasonal ⁵ and monthly ⁶ variation in blood pressure was first described in 1921 and 1930. Recent years’ focus on global climate change and impact on human health ⁷ and valuation of so-called local, traditional or indigenous knowledge ⁸ has renewed the interest in this area of research.

1.2 Seasonal variation – the burden of disease

Winter excess mortality is a worldwide phenomenon ⁹. In the European countries Spain, Italy, Portugal, Greece and the UK, the estimated total excess winter all-cause mortality is approximately 100,000 deaths, i.e. 16-28 % winter excess mortality ¹⁰, where the UK alone has 30,000-37,000 annual winter excess deaths, i.e. 18 % winter excess mortality ^{10, 11}. In Norway the corresponding number is 2,600 deaths or 12 % excess mortality in winter ¹². Winter excess mortality is mainly due to CVD ^{12, 13}. Various studies from Europe ^{14, 15}, USA ^16-19 and Oceania ^{20, 21} report the morbidity or mortality from MI to be 22-70 % greater in winter than in summer.

9 1.3 A systematic review

Systematic literature reviews were performed both before initiation of, as well as regularly during, this PhD work. This particular systematic review will be an attempt to be a scan of the English literature in this research area found in PubMed registered medical journals, from the period before this PhD work was carried out. A systematic review allows an objective appraisal in contrast to traditional narrative reviews, that are prone to bias and error ²². To meet the criterion for a systematic review ²³, a protocol was prepared before the literature search was performed, and consisted of a formulation of the review question, an a priori definition of eligibility criteria, and a plan for the comprehensive search and for assessment of the methodological quality of the studies.

The results will provide information on the extensiveness of the literature with a superficial review of possible shortcoming in existing knowledge without in-depth details on methodology or results for each study. The short discussion of the results will lead to a rationale for performing a new study in this field of research.

1.3.1 Objectives of the review

There were three objectives, which made the basis of three systematic reviews with two different exposures and two different outcomes:

To consider the effect of season on MI morbidity and/or mortality.

To consider the effect of temperature, wind, atmospheric pressure, humidity and snowfall on MI morbidity and/or mortality.

To consider the effect of season on systolic and diastolic blood pressure, heart rate, body weight, cholesterol, high-density lipoprotein (HDL) cholesterol, triglycerides, C-reactive protein (CRP) and fibrinogen.

1.3.2 The search strategy

Three main literature searches were performed. Medline via OvidSP was searched March 19, 2012 for the period from 1946 (Medline archive start) to end of 2007 (PhD project start) for the following U.S. National Library of Medicine Medical Subject Headings (MeSH) terms (based on the tree structure): “death, sudden, cardiac”, “acute coronary syndrome”, “coronary disease”, “coronary artery disease”, “myocardial infarction” AND “seasons” respectively “weather”, and for: “blood pressure”, “heart rate”, “body weight” , “cholesterol”, “triglycerides”, “cholesterol, HDL”, “C-

reactive protein”, “fibrinogen” AND “seasons”. All subheadings were included and the searches were limited to studies of humans, adults over 19 years, published in English.

10 1.3.3 Inclusion and exclusion criteria

Studies were included when MI, respectively the various CVD risk factors, were included as outcome, and season, respectively weather, was included as exposure. To retrieve information from studies of mortality suspected to origin from MI or from the period before MI was a commonly used concept, studies using the following terms as outcomes were included: morbidity or mortality from acute coronary syndrome (ACS), combinations of MI and angina pectoris (AP), coronary disease (CD) and coronary artery disease (CAD), including terms like “arteriosclerotic heart disease”, “coronary thrombosis”, “coronary artery occlusion” and “coronary insufficiency”, or mortality from ischemic heart disease (IHD), coronary heart disease (CHD), sudden cardiac death (SCD) or cardiac arrest (CA).

If several outcomes were studied, only results from analysis of MI are presented here. Results of sub- group analyses (gender, age, et cetera) are not presented here unless only results from sub-analyses were reported.

Studies were excluded if investigating seasonality of specific MI outcomes (S-T elevation versus non S-T elevation MI, fatal versus non-fatal MI, et cetera). Studies were also excluded if they were not original research articles (comments, letters, editorials or conference abstracts) or were a case report or review study, or did not conclude with the question of interest by reporting an association or non- association. In addition, studies were excluded if they, for the association with season, included one season analysis only, or, for the association with weather, included non-meteorological variables or other meteorological variables than temperature, wind, atmospheric pressure, humidity or

precipitation as snow. Studies were also excluded if they included mixes of meteorological variables (weather-types), only one extreme weather event, or were controlled experiments (cold-exposure in a laboratory setting or controlled diets in cross-over designs, et cetera). Studies reporting results for persons with seasonal affective disorder only were also excluded.

Reference lists for all included studies, as well as for identified review studies, were searched for any further studies, and also worked as a test of fit of the main search. This included an investigation of the search terms for all additional studies, to see if there were other search terms that should have been added to the main search. These references were processed in the same way as studies from the main search. Only references from 1946 and forward were included, in accordance with the main search.

1.3.4 Procedure

After the search, titles and abstracts were screened for relevance based on the inclusion and exclusion criteria. If the content of the title and abstract was not extensive enough to decide if the study should be included or excluded, or if the abstract were missing, a full-text version of the paper

11

was obtained and superficially reviewed for final decision. For studies where an interpretable title or abstract were missing (for decision of study inclusion or exclusion) and a full text version was not available in Medline, both a Google search and a Bibsys library search (to order the article via international library resources) were performed for retrieving of full text version.

Due to the extensiveness of the search of seasonal pattern in CVD risk factors, a modification was made for this search. Here, only studies available in full-text found in the main search were included, and reference lists were not searched for further studies. This means that both if an interpretable title or abstract were missing (for decision of study inclusion or exclusion) and for studies likely to be included based on the available abstract content; if a full text version was not available in Medline or Google, the full-text version was not ordered via library resources.

For each included study from the three searches, the full text version (either found as an electronic version in Medline, or, for the two first searches, obtained electronically via Google or ordered and received as print per mail via library resources) was scanned and the following information was obtained: study design, time-period or follow-up time, number of subjects included, study

population, geographical area, and definition of outcome and exposure, as well as a short description of the result of the study.

1.3.5 Results

A total of 74 studies meet the inclusion criteria for reviewing the association between season and MI (table 1) and 56 studies for the association between weather and MI (table 2). There are 22 studies reporting both seasonal and weather effects on MI, and these exist in both table 1 and table 2, reporting only the exposure of interest for the specific search. A total of 72 studies meet the inclusion criteria for reviewing the association between season and the various CVD risk factors (table 3).

1.3.5.1 Seasonal variation in myocardial infarction morbidity and mortality In 51 11, 14-21, 24-65

of 74 (69 %) studies there is a seasonal variation in MI morbidity or mortality, and from these, 30 11, 14-16, 18-21, 24-45

(59 %) studies report a winter peak only.

1.3.5.2 Effect of weather on myocardial infarction morbidity and mortality

Among the 56 studies of effect of weather on MI morbidity or mortality, an effect is found for temperature in 42 13-15, 17, 20, 34, 36, 40-42, 60, 61, 66-95 of 49 (86 %) studies, for wind in 2 ^{24, 41} of 4 (50 %), for atmospheric pressure in 5 24, 57, 66, 75, 96

of 16 (31 %), for humidity in 8 14, 41, 67, 69, 72, 77, 78, 84

of 17 (47 %), and for precipitation as snow in 6 24, 41, 80, 87, 97, 98

of 9 (67 %) studies. For those studies finding an effect, MI morbidity or mortality is positively associated with wind and snowfall, positively, inversely,

12

U- or V-shapely associated with atmospheric pressure and humidity, and generally (83 %) inversely associated with temperature, except from 7 studies of temperature and MI that report a positive ^70,

71, 74

or a U-shaped 14, 69, 95, 99

relationship.

1.3.5.3 Seasonal variation in cardiovascular disease risk factors

Several studies investigate the seasonal pattern in more than one risk factor. Seasonal variation is investigated in 32 studies for systolic blood pressure, 31 for diastolic blood pressure, 4 for heart rate, 12 for body weight, 27 for cholesterol, 23 for triglycerides, 19 for HDL cholesterol, 7 for CRP and 11 for fibrinogen.

Systolic blood pressure has a seasonal pattern in 28 ^100-127 (87.5 %) studies, and from these a winter peak is reported in 23 100-105, 108-113, 115-123, 125, 126

(82 %). Diastolic blood pressure has a seasonal pattern in 26 100, 101, 103-122, 124, 125, 127, 128

(84 %) studies, with a winter peak in 20 100, 101, 103-105, 108-113, 115-122, 125

(77 %). Heart rate has a seasonal pattern in 1 ¹⁰⁷ (25 %) study, peaking in summer. Body weight has a seasonal pattern in 7 111, 117, 129-133

(58 %) studies, all with a winter peak. Cholesterol has a seasonal pattern in 21 105, 106, 109, 115, 116, 127, 130-132, 134-145

(78 %) studies, and from these a winter peak is reported in 17 105, 109, 115, 116, 127, 130-132, 134-138, 142-145

(81 %). Triglycerides has a seasonal pattern in 14 105, 106, 109, 115, 116, 127, 131, 132, 134, 135, 139, 140, 145, 146

(61 %) studies, with a winter peak in 2 ^{106, 109} (14 %). HDL cholesterol has a seasonal pattern in 16 105, 106, 115, 116, 130-132, 134-136, 138, 139, 142-144, 147

(84 %) studies, with a winter peak in 11 105, 115, 116, 131, 132, 134-136, 138, 142, 144

(69 %). CRP has a seasonal pattern in 5 106, 142, 148-150

(71 %) studies, with winter peak reported in 2 ^{142, 148} (40 %). Fibrinogen has a seasonal pattern in 9 142, 148-155

(82 %) studies, and from these there is a winter peak in 6 142, 148-150, 152, 153

(67 %) studies.

1.3.5.4 Methodological considerations – the review

From the 108 studies investigating the effect of season and/or weather on MI 5 50, 156-159

(5 %) report first-ever MI. A total of 8 28, 43, 63, 72, 74, 75, 95, 156

(7 %) studies include separately reviewed cases with a description of a validation-of-event protocol with diagnostic criteria, while the remaining used routinely collected statistics from mortality registers or hospital discharge databases. There is a lack of information on time-period of study conduction in 8 15, 20, 64, 65, 74, 81, 88, 157

(7 %) studies, number of included subjects in 22 11, 13, 15, 16, 19, 21, 32, 36, 42, 70, 71, 76, 79-81, 85-87, 90, 92, 98, 160

(20 %) studies and

geographical area in 1 ¹⁵⁷ (1 %) study. A total of 35 11, 13, 14, 16, 17, 19, 21, 25, 26, 30-33, 36, 37, 41, 42, 46, 51, 69, 76, 79-82, 85-87, 90, 92, 98, 99, 109, 160, 161

(32 %) studies report mortality only. From the 73 studies reporting morbidity (included combinations of morbidity and mortality) 14 14, 20, 29, 43, 50, 60, 61, 68, 74, 75, 91, 156, 162, 163

(19%) claim to be population-based while the remaining studies include events from hospital admissions data only. In 5 53, 57, 157, 159, 164

(7 %) of the morbidity studies the information of data source lack or the available information is too limited to classify as population-based or not. From the 43 studies

13

reporting the seasonal distribution (i.e. not monthly only) of mortality or morbidity 20 15, 19, 28, 31, 32, 35, 39, 45, 46, 48, 49, 52, 54, 57, 60, 61, 65, 78, 83, 162

(46.5 %) do not define the seasons.

From the 72 studies investigating seasonal variation in CVD risk factors, 12 102, 103, 109, 118, 126, 127, 146, 147, 165-168

(17 %) claim to be population-based or partly population-based. In 3 105, 134, 153

(4 %) studies the information lack or is too limited to classify as population-based or not. A total of 16 102-106, 112, 124, 126, 127, 134, 141, 146, 147, 152, 154, 165

(22 %) studies use only one measurement per individual or the available information is too limited to classify as repeated measurements or not. A total of 16 122, 128, 137-140, 145, 148, 155, 168-174

(22 %) studies have fewer than 30 subjects, and 48 100, 101, 104, 107, 108, 110-117, 119-124, 127, 128, 130- 132, 136, 137, 142, 144, 146-152, 154, 165, 167-177

(67 %) studies report findings from populations of patients on certain treatment, very limited age-ranges or one sex only. There is a lack of information on time- period of study conduction in 19 100, 112, 115, 116, 120-122, 125, 129, 131, 132, 136, 138, 139, 148, 170, 171, 174, 178 (26 %) studies, number of subjects in 2 ^{154, 179} (3 %) studies and geographical area in 14 100, 107, 115, 116, 120, 121, 136, 148, 171, 173, 174, 176, 178, 179

(19 %) studies. From the 42 studies reporting the seasonal distribution (i.e.

not monthly only) of risk factors 17 100, 102, 108, 120, 121, 126, 127, 131, 132, 134, 136, 141, 143, 153, 165, 171, 175

(40.5 %) do not define the seasons.

1.3.5.5 Geographical differences – the review

By roughly dividing the studies into broad different geographical areas there are more studies reporting seasonal variation in MI with winter peaks in the UK 11, 15, 26, 30, 40, 42, 43, 45, 60, 61

, the Mediterranean countries 14, 27, 34, 48

, Southern hemisphere countries 20, 21, 29, 32, 33

and the USA as a whole 16, 18, 25, 38, 41, 44, 64, 65

. Inconsistency in results is found in reports from Canada 37, 42, 162

, Japan ^28,

35, 36, 47, 180, 181

and Southern USA ^{19, 62, 63}. Lack of seasonal variation in MI is more often reported from the Nordic countries 159, 163, 182-184

, Northern USA 18, 156, 161, 185

and countries close to the Equator ^{83, 186,}

187.

Similarly, studies from the UK 15, 40, 60, 61, 76, 77, 85, 86, 90-92, 94, 160

, the Mediterranean countries 34, 48, 67, 72

and Southern hemisphere countries 20, 79, 81, 88

more often report inverse relationships between MI events and temperature. Temperature is more often reported to be U-shapely associated with MI events in countries close to Equator ^{69, 99}. For studies reporting from several countries, areas with colder climates show less effect of temperature on MI events than areas with milder climates 13, 42, 68, 80. For risk factors, the geographical differences are not that evident, and studies from most areas report a seasonal pattern with a winter peak. However, in a study reporting from several countries, populations in areas with colder climates show less seasonal variation in systolic blood pressure than populations in milder climates ¹⁰². Similarly, studies from USA show larger seasonal variation in cholesterol levels in Southern compared to Northern areas ^{131, 132}.

14 1.3.5.6 Limitations of the review

From the two searches investigating seasonal variation or weather effects on MI, 46 studies were found in reference lists after the main search. From these, 26 had search terms matching the main searches, and why they were not found in the main search cannot be explained. The rest had other MeSH terms or other combinations of MeSH terms. From these, 3 studies had the term “myocardial infarction” combined with search terms other than “weather” (“meteorological concepts”, “hot temperature” or “climate”, or “atmospheric pressure” which was not included in the term

“weather”) and 4 studies had no term for “season” nor “weather”. In addition to this, among the older (i.e. from around 1960 and earlier) studies, 5 had only one MeSH term (“myocardium” or

“periodicity”) and 6 did not exist in Medline. Further, 1 study was incorrectly marked as Spanish even if it was an English version, and 1 study was incorrectly marked with the term “cerebrovascular disorders” even if this was not among the study endpoints. In total, from both main searches and reference searches, 27 studies were not available in full text versions in Medline or Google, and were ordered from the library via Bibsys.

From the search of studies investigating seasonal variation in CVD risk factors, 30 studies were not found in full text versions and therefore excluded. These were both studies that should have been included based on the available information in the title and abstract, and studies that lacked this information to evaluate whether to be included or not.

As described above the review is fairly extensive for the first two searches, but more limited for the search on risk factors. Most studies of interest were identified, and the investigation of search terms in studies found in references lists and not in the main searches did not reveal a large number of search terms that should have been added to the main search. There are, however, limitations for the review as a whole. First, only one database was searched. Further, there is a possibility of language bias, negative publication bias as well as limitations due to the time-frame (i.e. exclusion of studies before 1946, as these do not exist in Medline). However, the main limitation of this review is the lack of in-depth methodological considerations of the various studies, especially the lack of investigation of the statistical methods and a discussion of their quality and appropriateness. Even if a meta-analysis was not possible to perform due to the large variation in methods, some kind of comparison of effect, especially between the observed differences in different geographical areas, would have been of interest. An investigation of possible time trends in seasonal variation could have been an additional issue of interest.

15 1.3.6 Discussion with rationale for the thesis

The literature review shows that the main body of studies reports a seasonal pattern with winter peak in morbidity and mortality from MI and in levels of CVD risk factors, and that this has been studied extensively in the last half century. This calls for a rationale for this thesis. Epidemiological studies must progress from descriptive to analytical and further to studies of effectiveness, which leads to public health prevention programs, but there is a tendency in epidemiology to a stagnation in development of study design, leading to rediscover evidence from past studies – so-called circular epidemiology ¹⁸⁸.

The argument for a new study in this area of research is based on the limitations in previous study methods. Firstly, few studies are population-based, but are based on hospital discharge registers (for MI) or measurements within special subgroups (for risk factors), which can result in selection bias.

Excluding non-hospitalized MI events does not mirror the population numerator and thus does not account for all events in a community. A non-defined population (denominator) gives an unknown number of individuals at risk, and factors like seasonal migration may change the population during the year ¹⁸⁹. Secondly, few studies of MI are based on validated events, but use routine statistics, which can result in misclassifications. Many studies are based on mortality data only, and for those investigating morbidity, few distinguish between first and recurrent events, which may have different incidence patterns ¹⁹⁰ due to medical treatment that influence the physiological effects to seasonal exposure or patient advice that changes the behavioural pattern. Several studies of risk factors do not use repeated measurements in the same individual, which is important as individual and population seasonal patterns may differ ¹⁹¹. Further, few studies define the exposure season, which can give different results depending on how season is defined ^{192, 193}. Lastly, it seems to be different impacts of season and weather in different geographic and/or climatic areas, and few studies are conducted in high latitude and/or cold climate areas.

To perform circular epidemiology by identical replication is a stagnation ¹⁸⁸, but research with improved and innovative methods to minimize bias should be performed on relatively established research findings ¹⁹⁴. The limitations mentioned here could have been met by using unique personal identity numbers to search population-, disease- and death registers and thereby follow a defined general population of all ages and both sexes with person-time as denominator and well-validated endpoints as numerator. The rationale for this thesis is therefore the attempt to investigate seasonal and meteorological impact on the specific endpoint of validated first-ever MI and seasonal variation in repeated measurements of risk factors for this disease in a well-defined population in a subarctic area.

16 2. AIMS OF THE THESIS

The overall aim of the thesis is to describe the seasonal pattern in MI incidence and in CVD risk factors in a subarctic population. In order to fulfil the overall aim, three more specific objectives were formulated. These objectives made the basis for the data analysis in three different studies based on data from the subarctic cohort study the Tromsø Study;

To investigate the seasonal variation in first-ever non-fatal and fatal MI (paper 1).

To investigate the impact of daily meteorological variables on first-ever MI (paper 2).

To investigate the seasonal variation in CVD risk factors in repeated measurements (paper 3).

17 3. METHODS

3.1 Tromsø – location, climate and seasons

Tromsø is the regional centre of Northern Norway, situated 400 km north of the Arctic Circle at 69°39’N. The municipality has approximately 70,000 inhabitants. Tromsø has extreme seasonal variation in daylight. From approximately May 18 to July 26 the sun does not set, which gives two months of 24 hours daylight or so-called “midnight sun” during the summer. From approximately November 28 to January 14 the sun is below the horizon, which gives two months of dark winter season or so-called “polar night”. The climate is harsh and the weather is constantly changing, but because of the Gulf Stream, the climate is generally mild compared to other areas at the same latitude. Therefore, even if Tromsø is located above the Article Circle, the climate is subarctic as defined by the Köppen climate classification ¹⁹⁵. The city centre of Tromsø is situated on an island, with no industry causing air pollution. On certain days in autumn and spring there is a risk of increase in air pollution due to the winter use of studded tires on cars, but because of the local geographic with frequent replacements of air, accumulation of air pollution is not likely ¹⁹⁶.

The astronomical seasons are the same in Tromsø as for all geographical areas on the Northern hemisphere, where the spring (usually March 21) and autumn equinox (usually September 23) ¹⁹⁷ define the winter and summer season, with winter in the darkest months (November-January), summer in the lightest months (May-July) and spring (February-April, increasing daylight) and autumn (August-October, decreasing daylight) between these. The meteorological definitions of seasons depend on geographical area. Months with a daily mean temperature below 0°C is defined as winter, months with daily mean temperature above 10°C is defined as summer, and months with a daily mean temperature between 0°C and 10°C are defined as spring (increasing temperature) and autumn (decreasing temperature) ¹⁹⁷. For Tromsø, the meteorological definitions of seasons are as follows; winter: November 6-April 13, spring: April 14-June 22, summer: June 23- August 18, and autumn: August 19-November 5 ¹⁹⁷. For this thesis, data on meteorological variables from the Tromsø Weather Station were collected from the official Norwegian Meteorology Institute webpage Eklima ¹⁹⁸.

3.1.1 Meteorological data and definitions of seasons used in paper 1-3

In paper 1, the seasonal pattern in MI incidence was investigated in a Cosinor model based on time as month (month adjusted to 30.4375 day length) with one cycle per year. To describe the local climate, the monthly mean values of various meteorological variables were calculated for the 31 year period as a whole, based on monthly means from January 1974 to December 2004. These values also

18

estimated the seasons, based on the definitions described above. Whole months were used to model season, for both astronomical and meteorological season models. This is a simplified version of seasons than those described above, which use accurate dates. The astronomical season model divided the year into four three-months seasons depending on seasonal light (winter season as November-January). The meteorological season model divided the year into four seasons depending on monthly mean of daily mean temperatures (winter: November-March, spring: April-May, summer:

June-August, autumn: September-October). April was close to be included as a winter month, as the first day of spring is April 14, but still with more days of spring. June had a monthly mean of daily mean temperature of 9.2 °C, but was included as a summer month because the temperature value was closer to the summer month temperatures than to the spring month temperatures. For both the definitions of season, a model dividing the months into winter and non-winter (spring, summer and autumn) season was also included.

In paper 2, meteorological data for each day of follow-up were linked to person-years at risk and number of events. Based on assessment of possible time-lag phenomenon, a three-day average (the weather on the date of the event and the two previous days) of each meteorological variable was used as exposure variables. In addition to that, to describe the local climate, the yearly mean values of various daily meteorological variables was calculated for the 31 year period as a whole, based on daily values from February 14, 1974 to December 31, 2004, the total period of follow-up. The meteorological defined seasons using the accurate dates as described above was used to perform separate analysis for winter season for the weather variable precipitation as snow, to investigate if there was any difference in MI incidence after snowfall in winter compared to snowfall in other seasons.

In paper 3, the seasonal pattern in CVD risk factors was investigated in Cosinor models based on time as date of screening as the main exposure variable with one cycle per year. Here, meteorologically defined seasons using the accurate dates as described above were used only to describe the results.

Likewise, to describe the local climate the monthly mean values of various meteorological variables were calculated for the 30 year period as a whole based on monthly means from January 1979 to December 2008.

3.2 Study population – The Tromsø Study

The Tromsø Study is a single-centre, population-based, prospective health study conducted in the municipality of Tromsø. The Tromsø Study consists of six repeated health surveys (Tromsø 1: 1974, Tromsø 2: 1979–80, Tromsø 3: 1986–87, Tromsø 4: 1994–95, Tromsø 5: 2001–02 and Tromsø 6:

19

2007–08) including a second extended subsample-screening after the first visit in the three last surveys (Tromsø 4 visit 2, Tromsø 5 visit 2 and Tromsø 6 visit 2). Both total birth cohorts and random samples of Tromsø inhabitants were invited by written mail-sent invitations. The overall participation rate ranged from 66-85 % (when adjusting for deaths and emigration), and were even higher for the second visit screenings ¹⁹⁹. The Tromsø Study has been approved by the Norwegian data Inspectorate and recommended by the Regional Committee of Research Ethics. In Tromsø 4–6, the participants signed a written consent. Participants are free to withdraw their consent at any time and also to give new consent later, therefore the number of participants with valid consent may vary over time. An overview over the Tromsø Study sample is given in table 4, numbers of examinations attended is given in table 5 and a flowchart for the subsamples used in paper 1–3 is given in figure 1. Further information about the Tromsø Study, including the invitation letters, consent forms and

questionnaires for each survey is available at the study web pages www.tromsoundersokelsen.no ²⁰⁰. 3.2.1 Study population used in paper 1–3

Paper 1 is a follow-up study of MI incidence among all participants from the first five surveys (Tromsø 1-5: 1974–2002). They were followed from enrolment (first date of examination) to December 31, 2004. This gave a total of 38,164 participants, where 160 subjects were excluded because of emigration from Tromsø before start of follow-up, 222 because of non-consent and 390 because of prevalent MI, which resulted in 37,392 participants for inclusion in the analysis. New participants from the last survey (Tromsø 6: 2007–2008) could not be included, as end point validation was only complete until the end of 2004. Median length of follow up was 15.7 years.

Paper 2 is a follow-up study of MI incidence among all participants from the first five surveys (Tromsø 1-5: 1974–2002). They were followed from enrolment (first date of examination) to December 31, 2004. This gave a total of 38,164 participants, where 160 subjects were excluded because of emigration from Tromsø before start of follow-up, 222 because of non-consent, 390 because of prevalent MI and 5,282 because they were younger than 35 years at the end of follow-up, which resulted in 32,110 participants for inclusion in the analysis. To have complete age-groups throughout the follow-up period, subjects younger than 35 at the end of follow-up were excluded, as this age- group grew older without including new participants as years passed by. As for the study in paper 1, new participants from the last survey (Tromsø 6: 2007–2008) could not be included, as end point validation was only complete until the end of 2004. Median length of follow was 17 years.

Paper 3 is a study of repeated measurements of CVD risk factors in participants from the second to the last survey (Tromsø 2-6: 1979–2008), included visit 2 (a second and more extensive screening) in Tromsø 4-6. This gave a total of 39,059 participants, where 224 subjects were excluded because of

20

non-consent, 6 because of missing attendance date in 1979–80, 7 because of attendance without invitation and 785 because they were younger than 20 years at enrolment, which resulted in 38,037 participants for inclusion in the analysis. Subjects younger than 20 at enrolment were excluded because of participation in Tromsø 3 only – if they were invited again (older than 20 years), they were included. Measurements from the first survey were not included due to inconsistency in measurement methods compared to later surveys. For participants included, 41% had at least 3 or more examinations, and 28% had between 4 and 8 examinations with repeated risk factor

measurements. There were various numbers of measurements for each risk factor; from 92,641 measurements of cholesterol in 37,986 participants, to 25,421 measurements of fibrinogen in 16,450 participants, i.e. not all risk factors were measured in each survey and not all measurements were performed in all participants.

3.3 Myocardial infarction case identification and definition used in paper 1 and 2

A huge work to identify and define each case of MI among the Tromsø Study participants is being performed by the Tromsø Study endpoint committee. To identify hospitalized cases of fatal or non- fatal MI among study participants the discharge diagnosis register at the University Hospital of North Norway (UNN), the only local hospital serving the Tromsø population, was searched. To identify out- of-hospital cases of fatal MI the study participant list was linked with the Norwegian Causes of Death Register. The unique personal identification system in Norway makes exact matching in register sources possible. For event ascertainment the endpoint committee followed a detailed protocol and examined all available medical records (including medical records from other hospitals and pre- hospital records from ambulance service, general practitioners, nursing homes and death

certificates). Among the cases 4 % were observed at another hospital than UNN, and of these 25 % were observed at a hospital in Northern Norway. Diagnostic criteria were based on clinical

presentation, electrocardiograms, levels of myocardial biomarkers, echocardiograms, results from angiography and/or autopsy. Fatal MI was defined both as death on the same date as the event and death within 28 days.

3.4 Cardiovascular disease risk factors measurements used in paper 3

Each survey followed a standardized examination protocol, and the methods used for physical examination (blood pressure, measurements of body weight and heart rate) and blood sampling were almost identical for all surveys. For blood pressure, there were some inconsistencies in measurement methods. In Tromsø 3-6 blood pressure and heart rate were measured with an automatic device, while in Tromsø 2 blood pressure was measured manually with a mercury

21

sphygmomanometer. Previous validated methods ²⁰¹ were used to transform the recordings from the automatic device, to adjust for this change in measurement method. The mean of the final 2 of the 3 readings in Tromsø 3-6 and the 2 readings in Tromsø 2 were used in the analysis. Weight was

measured with subjects wearing light clothes and no shoes. Blood samples were analysed (for total cholesterol, triglycerides, HDL cholesterol, CRP and fibrinogen) by standard methods at the

Department of Laboratory Medicine at the University Hospital of Northern Norway (names have changed for both department and hospital during survey period). Participants were not requested to fast. All data collection was performed by trained personnel.

3.5 Data from questionnaires used in paper 1

Self-reported data on smoking for Tromsø 1-5 was collected from questionnaires²⁰⁰ for purpose of stratified analyses of smoking status (current smoker or non-smoker) in paper 1. Definition was made by reported smoking status in the latest questionnaire available, and for cases, the latest available questionnaire before the event.

3.6 Statistical analyses 3.6.1 Cosinor

In paper 1 and 3 a sinusoidal seasonal pattern was fitted using cosine and sine terms to monthly incidence (paper 1) or date of screening (paper 3), known as the Cosinor procedure. The Cosinor analysis was developed by Halberg, Tong and Johnson ²⁰² and determines how much of the seasonal variation can be explained by a sinusoidal curve, a common method for analysing seasonality in health data ¹⁹¹. Cosinor analysis has the assumption that a cycle exists and that the cycle length is known ²⁰³. Using the Cosinor procedure helps determine the characteristics of this cycle; the

acrophase or phase (the time point at which maximum values occur, i.e. peak time point), the mesor (the mean of the fitted curve) and the amplitude of the wave form (the peak to mesor difference, i.e.

the size of the seasonal change from the seasonal mean) ²⁰³. A linear regression model using the Cosinor procedure can be described by the following equation:

y= β0+β1cos(2π t/P)+β2sin(2π t/P) + e

where y is the dependent variable of interest (the outcome variable), βs are regression coefficients, t is the time point, P is length of period, cos is cosine, sin is sine and e is the residual variation term. A Cosinor curve with parameters as described above is shown in figure 1.

22 3.6.2 Statistical models used in paper 1-3

In paper 1, data was arranged as survival time data with month of the year as time unit, where each subject was followed month by month (records of the data file were months of follow-up).

Seasonality in incidence was assessed with the Cosinor procedure in a Poisson regression model using the following equation:

log µ=log(t)+β0+β1cos(month*2π/12)+β2sin(month*2π/12)

where µ is the expected incidence rate, βs are regression coefficients, t is length of observation time in month for each record and month is the month under observation ranging from 1 to 12. In

addition to the Cosinor analysis, four different Poisson models were used to assess seasonal patterns.

These were two four-season models defined earlier (an astronomical and a meteorological season model) and, for both of these, a winter and non-winter model.

In paper 2, association between MI incidence and meteorological variables were assessed with Poisson regression. The meteorological variables were modelled using fractional polynomials. The records of the data file were dates of follow-up. When temperature (temp) was the independent variable, the best fitting fractional polynomials of degree two gave powers equal to (-2, -2). The following model was used:

log µ=β0+β1*temp^-2+β2*temp^-2*log(temp)

where µ is the expected incidence rate and βs are regression coefficients for temperature. The fractional polynomial models were also used to assess the relative risk of MI between upper and lower limits of the distribution of each meteorological variable.

In order to assess possible effects of unusual weather (unusual for the season), z-scores were

calculated for all meteorological variables (except for snowfall that seldom occur in summer) for each week of the year. The association between the z-scores and MI were assessed in similar fractional polynomial models as for the models above.

A linear model was used to estimate the percentage change in risk of MI per standard deviation increase in each meteorological variable. For the variable temperature (temp) the following model was used:

log µ= β0+β1temp

In paper 3, data was arranged with one record for each observation. The seasonality in the risk factors was assessed using a linear mixed model. Regression analysis was fitted by restricted

23

maximum likelihood. The models were specified as random coefficients or multilevel models

(thereby controlling for repeated measurements). Two models were used. For the population model, time unit was day of the year, with cosine and sine functions as fixed effects and a random intercept for each individual. For the individual model random terms of the cosine and sine effects were added. Both models included fixed effects of age and sex.

The population model had the following equation:

y= β0+β1X1+β2X2+ β3age+β4sex+β5time+ 

 7 1

i λisurveyi +u+e The individual model had the following equation:

y= β0+β1X1+β2X2+ β3age+β4sex+β5time+ 

 7 1

i λisurveyi +u+V1+V2+ e

where y is the risk factor (the outcome variable), X1, X2, age, sex, time, and surveyi are fixed effects independent variables, where X1 = cos(day of year*2π/365.25), X2 = sin(day of year*2π/365.25), surveyi are indicator variables of survey number, βs and λis are regression coefficients, u is the random intercept, V1 and V2 are random effects for the Cosinor variables X1 and X2 and e is the residual variation term.

For all papers two-sided p-values < 0.05 were considered statistically significant. Analyses were performed using stratifications by sex (paper 1, 2 and 3), age (paper 1 and 2), smoking status (paper 1), and/or models included adjustment for sex and age (paper 3), season (paper 2) or time as survey number (paper 3). Effect modification was assessed for the variables time as calendar year (paper 1) or calendar days (paper 2), sex (paper 1 and 2), age (paper 1 and 2) and smoking status (paper 1) by including cross-product terms for the effect modifier with the two Cosinor functions of month (paper 1) or with each meteorological variable (paper 2). Assessment of circadian and weekly variation in examination time and association with month of year was investigated descriptively (paper 3).

Analyses was performed using STATA (Stata Corp LP) version 10 (paper 1) and 11 (paper 2), and R (www.r-project.org) version 2.13.2 using the lme4, circular and season packages (paper 3).

In paper 1, the Wald test was used to assess seasonality (the seasonal terms; sine and cosine) in the Poisson model. The likelihood ratio test was used to assess the effect of season when season was categorized in 4 levels. A power calculation was performed to assess the possibility to detect a seasonal pattern. Given a sample size of 37,392 subjects and an incidence rate of 2.7 per 1,000 there

24

was an 80% power to detect a rate ratio of 1.27 change in amplitude from the mesor in the Cosinor model.

In paper 2, the likelihood ratio test was used to assess the association between MI incidence and each meteorological variable expressed as fractional polynomial terms (untransformed and as week- specific z-scores). Confidence intervals were used to assess the impact on MI of each meteorological variable expressed as linear terms.

In paper 3, the likelihood ratio test was used to assess seasonality by comparing models with and without seasonal terms. This tests compared the model with or without seasonal terms as a fixed effect (test of seasonality on the population level) or as a random effects (test of seasonality on the individual level).

25 4. RESULTS – SUMMARY OF PAPERS

4.1 Paper 1

Seasonal variation in incidence in acute myocardial infarction in a subarctic population: The Tromsø Study 1974–2004

The aim of the study was to investigate the monthly and seasonal variation in first-ever non-fatal and fatal MI in the Tromsø Study cohort, with season defined by light and by temperature.

Among the 37,392 cohort members followed between 1974 and 2004, a total of 1,893 incident MIs were registered and analysed for seasonality.

The lowest incidence (fatal and non-fatal MI combined) occurred in July (2.88 per 1,000 person- years) and the highest in December (3.63 per 1,000 person-years). There was an 11% increase in risk of MI during the darkest winter months (November-January) compared with non-winter seasons.

Other seasonal models did not show significant seasonality in MI incidence.

There was no statistically significant interaction between monthly variation in MI incidence and time period, sex, age or smoking status.

The conclusion was that there was little seasonal variation in incident MI events in this subarctic population.

4.2 Paper 2

The effect of daily weather conditions on myocardial infarction incidence in a subarctic population:

The Tromsø Study 1974–2004

The aim of the study was to investigate the impact of daily meteorological factors on first-ever MI in the Tromsø Study cohort.

Among the 32,110 cohort members followed between 1974 and 2004, a total of 1,882 incident MIs were registered and linked with daily weather variables.

In the total population, wind speed had an inverse U-shaped (p=0.004) and humidity a U-shaped (p=0.004) association with MI incidence. MI incidence was not affected by changes in atmospheric pressure. There was an increase in risk of first-ever MI with decreasing temperatures (p=0.016) and with increasing snowfall (p=0.030) among individuals aged 65 years or older. In this age-group, the risk increased by 47% when comparing days with -10°C to days with 20°C, and by 44% comparing winter days with 10 mm snowfall with winter days without snowfall. Among those aged 35-64 there was no association with temperature, while snowfall showed a borderline significant inverse

26

association with MI (p=0.048). In women, a positive association was observed between MI and snowfall (p=0.035), while no association was observed in men.

There was no statistically significant interaction between the meteorological variables and time period, sex or age, except for snowfall and age (p= 0.0057), and snowfall and sex (p=0.023). Results were similar when including month as an adjustment for season.

The conclusion was that in this subarctic population the MI incidence was little affected by the weather, although older people should take extra precautions at cold temperatures and after heavy snowfall.

4.3 Paper 3

Seasonal variation in cardiovascular disease risk factors in a subarctic population: The Tromsø Study 1979–2008

The aim of the study was to investigate the seasonal variation in CVD risk factors in repeated measurements among the participants in the Tromsø Study.

From the 38,037 study participants examined in up to 8 screenings from 1979 to 2008, the seasonal pattern in 7 traditional CVD risk factors and 2 acute phase proteins associated with CVD was analysed.

Systolic and diastolic blood pressure, heart rate, body weight, total cholesterol and HDL cholesterol peaked in winter while triglycerides peaked in autumn, and CRP and fibrinogen peaked in spring. The seasonal pattern was significant for all risk factors (p<0.001). Maximum seasonal variation was 2 mmHg for systolic and 1 mmHg for diastolic blood pressure, 1.5 beats per minute for heart rate, 1 kg for body weight, 0.26 mmol/L for total cholesterol, 6 % for triglycerides, 0.062 mmol/L for HDL cholesterol, 12.8 % for CRP and 0.11 g/L for fibrinogen.

Results were similar in analyses stratified by sex. No association was found between month of year and day of week or time of day for the examination time.

The conclusion was that the seasonal variation in risk factors was highly statistically significant, but clinically likely too small to cause a seasonal pattern in CVD events in this subarctic population.

27 5. DISCUSSION – METHODOLOGY

The discussion of methodology is divided in two, firstly an overview over general methodological considerations and secondly considerations concerning each paper.

5.1 Overall considerations of internal validity 5.1.1 Study design

The Tromsø Study is defined as a prospective cohort study, as individuals are screened for risk factors prior to disease and followed up by repeated screenings. However, the design of the present studies was not pre-specified at the start of follow-up. Therefore the design can be called prospective non- concurrent data collection of exposure and outcome, as both the exposure (season or weather) and outcome status (MI or CVD risk factors) is not assembled during follow-up, but later. For paper 3, a short discussion of the appropriateness of the study design is included in the section about paper 3.

5.1.2 Study population – selection bias, non-response bias and loss to follow-up

Participation in the Tromsø Study is voluntarily and there is a possibility for selection bias, although the general participation rate is high (around 80% for the first five surveys, 65% for the latest survey, table 4). Differences in motivation (for example little interest in health screenings, which could give lower participation rates among younger individuals - a trend in the Tromsø Study ¹⁹⁹ as well as in other population studies in Norway and internationally) or other obstacles (for example physical impediment among older, chronic ill or bedridden patients) to participate can produce a non-

response bias among certain subgroups. Age should be controlled for here, either by stratifying, or by adjusting for, or by assessing the interaction. A lack of inclusion of all age-groups would limit the generalizability but not alter the results on seasonal distribution of disease. Differently, the lack of inclusion of for example homeless people might alter the results as they could be more exposed to seasonal meteorological factors. However, in Norway there is generally high socio-economic equality

204 and a welfare system covering all citizens²⁰⁵, and few individuals live on the street. Loss to follow- up is not of a major concern in the Tromsø Study, quite contradictory; the attendance rate in repetitive surveys is high among previous participants ¹⁹⁹. However, migration may create some loss to follow-up.

5.1.3 Measurement of outcome – misclassification bias

For paper 1 and 2, the use of MI as an endpoint gives possibilities for misclassification. In spite of thorough efforts to perform an accurate validation of each possible case, the decision lies with the researcher investigating the available information. Despite a strict protocol and researcher expertise, the final judgment is a subjective matter and therefore measurement errors like misclassification (or